Wechat Article Forge

# wechat-article-writer

> 从选题到发布的公众号一体化写作工作流

Multi-agent pipeline: Orchestrator delegates writing and reviewing to independent subagents. The orchestrator never writes or reviews — it routes, tracks versions, and enforces quality gates.

## Setup

```bash
bash <skill-dir>/scripts/setup.sh <workspace-dir>
```

Installs: bun runtime, bundled baoyu renderer deps, and a persistent preview server (`wechat-preview.service`, port 8898, auto-restart).

## Scope

**Handles:** Topic research → Chinese-first writing → quality review → scrapbook illustrations → WeChat formatting → publishing to WeChat draft box (via Official Account API or CDP browser automation).

**Does NOT handle:** Git/version control, non-WeChat platforms, post-publish analytics, WeChat messaging/customer service.

**Ends at:** Article saved to WeChat draft box. User publishes manually.

---

## Commands

Trigger any command below, or see `skill.yml` for the full trigger pattern list.

| Command | What it does |
|---------|-------------|
| `forge topic X` | Research trending angles, propose 3 options with hooks |
| `forge write X` | Full pipeline: research → publish (9 steps) |
| `forge draft X` | Write + format only, stop before illustrations/publish (steps 1-7) |
| `forge publish <slug>` | Publish an existing draft to WeChat |
| `forge preview <slug>` | Render preview, run format quality checks |
| `forge voice train` | Analyze past articles to extract voice profile |
| `forge status` | Show pipeline status and pending drafts |

If no subject given, loads from `session.json` (set by `forge topic`). See `references/data-layout.md`.

---

## Pipeline (9 Steps)

State persists to `pipeline-state.json` — survives compaction. See `references/pipeline-state.md`.

| # | Step | Who | Details |
|---|------|-----|---------|
| 1 | **Research + Prep** | Orchestrator | (a) `web_search` for topic angles + 5-8 sources. (b) Verify each source exists (fetch title/authors/venue). Save as `sources.json`...

# wechat-article-writer

> 从选题到发布的公众号一体化写作工作流

An OpenClaw skill that automates the full lifecycle of a WeChat Official Account (微信公众号) article: topic research, multi-agent writing with quality gates, scrapbook-style illustrations, formatting, and publishing to the draft box.

**v2.4.0** · Three publishing paths: Official Account API (recommended), browser tool, or direct CDP · Bundled baoyu renderer (no external dependencies)

---

## What It Does

```
forge write 关于AI编程工具的深度评测
```

That single command runs a **9-step pipeline**:

1. **Research + Prep** — topic angles, verified `sources.json` bank, voice profile, outline
2. **Write** via Writer subagent (Chinese-first, constrained to source bank)
3. **Review** via Reviewer subagent (blind 8-dimension craft scoring)
4. **Revise** — 4a: up to 2 automated cycles; 4b: human-in-the-loop if needed
5. **Fact-check** via Fact-Checker subagent (verify every claim, generate reference list)
6. **Format** to WeChat HTML via bundled baoyu renderer (`scripts/format.sh`, default classic theme)
7. **Preview** on persistent HTTP server at port 8898 for human approval
8. **Illustrate + Embed** — scrapbook images via article-illustrator, upload to WeChat CDN
9. **Publish** to WeChat draft box — via API (preferred) or browser automation (fallback)

---

## Architecture

```
Orchestrator (Main Agent) — routes, tracks, enforces gates
    ├── Writer Subagent — drafts + revises (Opus model)
    ├── Reviewer Subagent — blind craft scoring (Sonnet model)
    ├── Fact-Checker Subagent — verifies claims via web search (Sonnet model)
    └── article-illustrator — scrapbook images (after text passes)
```

- **Writer never self-reviews.** Constrained to verified source bank — must mark `[UNVERIFIED]` for anything outside it.
- **Reviewer is blind.** Never sees outline or brief — judges craft as a reader would.
- **Fact-Checker is independent.** Verifies every claim via web search, generates reference list.
- **I...